Design and develop trajectory-based stream storage and query processing techniques to support real-time trajectory exploration and mining algorithms.
Acting as the common back-bone to support the vast variety of trajectory mining challenges stemming from diverse use case scenarios as addressed in WP-1, WP-3, WP-5, WP-8 this component of the Time-Trail Vault advances stream database processing towards spatiotemporal events. The variety of data types and the complexity of spatiotemporal pattern detection call for innovations in the lower parts of the software architectures, i.e., the core of the database engine that powers the data mining system. In particular, it should deal with partial time series possible stored in auxiliary file-repositories.
Description of work
The driving applications are mostly the trajectory streams from TomTom and KNMI. They both call for innovative techniques to handle both bulk loads, enable short circuit responses, and integrate seamlessly with large trajectory repositories. Novel data exploration techniques have to be developed to meet the special characteristics of data that is "only passing by". The challenges range from fast outlier detection, on-line clustering and aggregation of data streams to continuous analysis of the data stream(s) to derive information "on-the-fly" and support instant decisions .
Furthermore, existing stream processing systems do not provide the flexibility and processing capacities to efficiently implement and perform complex data mining tasks on high-volume data streams. The envisioned solution is to leverage the flexibility and scalability of existing high-performance database technology. Enriching it with novel efficient stream-mining primitive’s yields the highly scalable back-end support for a large-scale high-performance stream mining architecture.